the coding is really awesome . i love it.. please would you provide the rest of algorithms of machine learning like adaboost, XGBoost etc and also deep learning ...
Probably need to learn the math beforehand, you can't just go into this without studying the math for a while. It takes time and a 14 minute video will never cut it.
"then we can simplify this a little bit so we can first get rid of p of x because this depends not on y at all so just throw this away " What?? That's then totally not the same formula, no clue what's happening here. " all these probabilities here are values between 0 and 1 and if we multiply this then the number can become very small and we can run into inaccuracies so for this we apply a little trick instead of the product we do a sum and then we apply the logarithm so if you apply the logarithm we can change the product with a sum and then this is the final formula to get y" , Ma man u just created a completely different formula and didn't provide any actual explanations or even maths references to check. this is heavily poorly explained and rushed. :/
I agree but as an explanation for the first part we get rid of P(X) because for all possible values of classes (y) the X(the test sample) remains the same. Also for the logarithm part log(a.b) = log(a) + log(b) and as the probabilities will be between 0 and 1 the value of log(p(x_n | y)) will be negative. Since we are only doing the argmax i.e. the value of y for which the formula gives the maximum output the formula more or less does the same thing. Note: I am just saying this will happen when we apply argmax, you are 100% correct in saying that the formula has been changed and there should be a better explanation provided for the same.
The goal is find the class with the highest posterior. P(X) will have little/no impact on determining which class will have a larger posterior so it can be removed. The product is converted to a sum of logs by a property of logarithms called the logarithmic product identity. This avoids the issue of numerical underflow--if you multiply a long list of floats, the product will converge towards 0. Using this property of logarithms allows us to avoid that issue. The video creator is assuming some familiarity with math, stats, and probability, but these are foundational to ML so its a fair assumption to make. I'd definitely recommend starting with those topics before tackling ML, or if you've studied them in the past, review any math you don't understand along the way. Actually going through and coding it yourself will also help a lot. Even though the solution is given, it'll give you the opportunity to modify the code to dive deeper into anything you don't understand, review those concepts, and develop a stronger understanding. I would recommend using a notebook environment so that you can write notes and formulas as well. That'll give you a resource to look back on if you need to review it in the future. Happy learning!
What recording software do you use?
the coding is really awesome . i love it.. please would you provide the rest of algorithms of machine learning like adaboost, XGBoost etc and also deep learning ...
Great job
thank you!
Excellent
And what about head on over to implement kalman filter or particle filter from scratch ? Lets take forex data:)
This won’t work on imbalance dataset. only class dominate which has more value, because mean and variance would be always higher the other class
We need to apply techniques to handle imbalance dataset
This video is way too fast and not at all explained
Probably need to learn the math beforehand, you can't just go into this without studying the math for a while. It takes time and a 14 minute video will never cut it.
"then we can simplify this a little bit so we can first get rid of p of x because this depends not on y at all so just throw this away " What?? That's then totally not the same formula, no clue what's happening here.
" all these probabilities here are values between 0 and 1 and if we multiply this then the number can become very small and we can run into inaccuracies so for this we apply a little trick instead of the product we do a sum and then we apply the logarithm so if you apply the logarithm we can change the product with a sum and then this is the final formula to get y" , Ma man u just created a completely different formula and didn't provide any actual explanations or even maths references to check.
this is heavily poorly explained and rushed. :/
I agree but as an explanation for the first part we get rid of P(X) because for all possible values of classes (y) the X(the test sample) remains the same.
Also for the logarithm part log(a.b) = log(a) + log(b) and as the probabilities will be between 0 and 1 the value of log(p(x_n | y)) will be negative. Since we are only doing the argmax i.e. the value of y for which the formula gives the maximum output the formula more or less does the same thing.
Note: I am just saying this will happen when we apply argmax, you are 100% correct in saying that the formula has been changed and there should be a better explanation provided for the same.
The goal is find the class with the highest posterior. P(X) will have little/no impact on determining which class will have a larger posterior so it can be removed.
The product is converted to a sum of logs by a property of logarithms called the logarithmic product identity. This avoids the issue of numerical underflow--if you multiply a long list of floats, the product will converge towards 0. Using this property of logarithms allows us to avoid that issue.
The video creator is assuming some familiarity with math, stats, and probability, but these are foundational to ML so its a fair assumption to make. I'd definitely recommend starting with those topics before tackling ML, or if you've studied them in the past, review any math you don't understand along the way.
Actually going through and coding it yourself will also help a lot. Even though the solution is given, it'll give you the opportunity to modify the code to dive deeper into anything you don't understand, review those concepts, and develop a stronger understanding. I would recommend using a notebook environment so that you can write notes and formulas as well. That'll give you a resource to look back on if you need to review it in the future. Happy learning!
I agree that the math part is rushed but u can find it on other resources. Look for explanations of "maximum a posteriori estimation"